The present application relates generally to an improved data processing apparatus and method and more specifically to mechanisms for controlling dynamic voltage and frequency scaling (DVFS) for simultaneous multi-threading (SMT) processors.
Dynamic Voltage and Frequency Scaling (DVFS) is a common technique for managing power. Most recently DVFS has been applied in the context of server and new benchmarks for power/performance efficiency. The challenge in using DVFS is determining what frequency to use, since the frequency determines the voltages to be used. Lowering of frequency and thus voltage results in less power consumption. The usual relationship between frequency/voltage and power consumed is non-linear with a significant benefit from being able to operate at lower frequency and voltage in terms of energy efficiency for the work to be completed. However, the complexity of new generation processors makes older utilization based techniques overly conservative in detecting slack in computing systems. Detecting slack may provide guidance on when frequency can be lowered to improve energy efficiency and still meet required throughput for a running application.